JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 90, NO.

B8, PAGES 6737-6753, JULY 10, 1985

RecurrenceRates and Probability Estimatesfor the New Madrid SeismicZone

ARCH C. JOHNSTON AND SUSAN J. NAVA

TennesseeEarthquake Information Center, Memphis State University, Tennessee

An updated frequency-magnitude relation for the New Madrid seismic zone is used to derive con-
ditional probabilitiesfor future, large New Madrid earthquakes.We estimate that there is a 40-63%
probabilityof an mb > 6.0(Ms > 6.3) eventoccurringby the year 2000 and an 86-97% probability by the
year 2035. The estimatesfor a great 1812-type event (M s > 8.3) are less than 1% probability by 2000
A.D. and lessthan 4% by 2035 A.D. These probabilitiesare contingenton many factors,a number of
which remain assumptionsbecauseof the lack of a geologicalor palcoseismological chronologyof past
New Madridactivity.A conditional
probability
requires
k.nowledge
of a meanrecurrence
time,thetype
of distribution, and the standard deviation of actual repeat times about this mean. Four assumed
distribution functions(Gaussian,lognormal, Weibull, and Poisson)were fit to recurrenceestimatesbased
on a combinationof historicaland instrumentalseismicitydata. Standard deviation was allowed to vary
betweenone third and one half of the mean recurrencetime, and a rangeof conditionalprobabilitieswas
generatedfor time intervals of 15 and 50 years from the year 1985. The largest uncertainty in this
procedure was the size of the seismicsource area to use for recurrence estimation. Calculations were
done for both a large and'a small sourcezone which led to variation in estimatedrecurrenceintervalsby
a factor of 2. The large sourcezone was favoredfor the final probabilityestimatesbecauseof the large
crustalvolumerequiredto storeelasticallythe strain energyfor great New Madrid earthquakes.

INTRODUCTION seismicitylistings,and (3) geologicaland/or tectonicevidence.

The New Madrid seismiczone is best defined geographi- The applicabletime frame for thesesourcesis decades,centur-
cally by the concentrationof earthquakeepicentersoccurring ies, and the Quaternary Period, respectively.Instrumental
in southeast Missouri, northwest Tennessee,and northeast Ar- seismicityis generally the most complete and homogeneous
kansas(Figure 1) and located by the Saint Louis University data set, but it suffersthe severelimitation of sampling only a
seismicnetwork [Stauder et al., 1974-1983]. Geologically, the small fraction of the repeat time of large events.Historical
zone lies within the Mississippiembayment, a sediment-filled seismicitymay span at least one recurrenceinterval of major
structural trough opening south to the Gulf Coastal Plain. quakesbut is usuallya more suspectdata set becauseof lack
Geophysically,it is containedwithin the Reelfoot rift [Ervin of completenessand rather large uncertaintiesin magnitude
and McGinnis, 1975; Braile et al., 1982], thought to have assignmentand epicentrallocation. Where recoverable,slip
formed in late Precambrian times and to be currently experi- rates on Quaternary faults have been incorporated into seis-
encing compressivereactivation of some structural features. mic hazardanalysisto goodeffect,e.g.,the Wasatchfault zone
Becausethe zone generateda seriesof great earthquakesin [Swan et al., 1980; Cluff et al., 1980] and intraplate Japan
the winter of 1811-1812, because it is located in a tectonic [Wesnouskyet al., 1982, 1983]. The ratio of seismicto aseis-
feature of dimensionscapable of supportinglarge earthquakes mic movementfor thesefaults is generallyunknown, and this
[Nuttli, 1983], and becauseit exhibitsa relativelyhigh level of introducesan uncertainty in this technique.
present-dayseismicactivity, it is consideredthe highest seis- Information on more recentfault movementsmay be gained
mic hazard zone in the United Stateseast of the Rocky Moun- by analyzing deformation preservedin sedimentsover and
tains [Applied TechnologyCouncil, 1978; Algermissenet al., along fault zones,as has been done in southernCalifornia
1982]. [Sieh, 1978; Raleigh et al., 1982], New Madrid [Russ, 1979],
Over the past decade,remarkable progresshas been made and Charleston, South Carolina [Obermeieret al., 1985]. Such
in our understandingof the geological,geophysical,and tec- estimatesare subjectto their own set of uncertaintiessuch as
tonic framework of the New Madrid seismiczone. Many as- the difficultyof establishingreliabledatesof the deformational
pectsof the seismicity(suchas spatialdimensionsof the active episodesand of determining magnitudes of the causative
portions of the rift, style and depth of faulting, and estimates events.
of sourceparametersand groundmotion) are today subjectto No first-hand fault slip information is available for the New
fairly well-definedconstraintscomparedto a decadeago. Such Madrid seismic zone because of the thick cover of alluvium.
progresshas not been matched, however,in the important Moreover, no chronology of past large slip eventsexists,al-
question of providing accurate estimatesof the recurrence though the conclusionof Russ [1979] that three such events
timesand associatedprobabilitiesof large New Madrid earth- occurred in a maximum interval of 2250 years provides a
quakes.In this paperwe will reviewand update the past work generaland important constraint.Thereforein this paper we
on this topic. In addition, we will use the derived recurrence must rely solelyon instrumentaland historicalseismicitydata
rates in combination with some assumedproperties of the for the estimation of New Madrid recurrence intervals. As will
seismicityto generateprobabilisticestimatesof future large be shown later, such information is not adequate to assign
New Madrid events. confidentlyprobabilitiesto theserecurrencevalues.In the ab-
Input data for estimatingrecurrenceintervalsare available senceof measuredvalues of suchnecessaryquantitiesas strain
from three sources:(1) instrumentalmonitoring, (2) historical rate or the distribution of actual recurrence times about the
estimated mean return periods, plausible values are taken
Copyright 1985 by the AmericanGeophysicalUnion. from theoretical studies or from literature on other fault zones

Paper number 4B5322. and are used to constructrough estimatesof the conditional
0148-0227/85/004B- 5322505.00 probability of large New Madrid earthquakes.These are the
6737
6738 JOHNSTONAND NAVA' NEW MADRID PROBABILITIES

II II II II II II II • II • II II II II II

• Z Z Z Z • Z o
u o
u Z Z
 JOHNSTONAND NAVA' NEW MADRID PROBABILITIES
91ø 90 ø 89 ø
IL
M0
ß
o
D41 e
. KY
. +
TN
AR
ß
!974- 1983
Fig. 1. Map of the central United Stateswith the 1974-1983 in-
strumentalseismicitydata set [Stauderet al., 1974-1983]. The bound-
aries of the two sourcezones used for frequency-magnitudedetermi-
nation are large zone, 35.0ø-39.0øN/87.7ø-91.5øW;small zone (the
New Madrid seismiczone), 35.0ø-37.0øN/89.0ø-90.5øW.
first published conditional probabilities for the New Madrid
zone. We hope that they will serve as a referenceframe that
can be refined and improved as additional data becomeavail-
able.
FREQUENCY-MAGNITUDE RELATION
The magnitude distribution of earthquake occurrencesis
generally observedto follow the Gutenberg-Richterrelation-
ship
log (No) = a - bM
where Nc is the number of events greater than or equal to
magnitudeM and a and b are constants.Applied over a given
time period, the number of earthquakesin a given magnitude
range is obtained,yielding a characteristicrecurrencetime for
that magnitude based on the assumption that the processis
stochastic(i.e., that the occurrencerate is not a function of
time). The constanta, the activity parameter, providesa mea-
sure of the overall occurrencerate of earthquakesin the zone
consideredand is the zero magnitude intercept on a semilog
plot. The slopeb, or b value, is controlled by the distribution
of events between the higher- and lower-magnitude ranges.
Only magnitudedistributionswith constantb value slopewill
be consideredin this study. Both constantsinfluencederived
recurrence times; therefore it is essential that the values and
the variance of theseconstantsbe determined as accurately as
possible.
Previous Studies
Early efforts at obtaining a recurrencerelation for the New
Madrid region used historical compilations and modified
Mercalli intensities rather than magnitudes [McClain and
Myers, 1970; Algermissen,1972; Mann et al., 1974]. The
6739
88 ø averagereturn period for large earthquakes(MMI o > X) as
• 59 ø
determinedby thesestudiesvariesbetween175 and 700 years
[Nuttli, 1974].
Table 1 provides a summary of previousNew Madrid re-
currenceformulas based on magnitude.These resultsare not
directly comparable becauseof differencesin magnitude, area
38 ø normalization, cumulative versus noncumulative number of
events,and varying time windows.They do, however,provide
an indication of general recurrence ranges developed by a
variety of workers using a variety of approachesand bound-
ary conditions. These data imply that a destructive New
Madrid earthquake (mb6.0)can be expected of the order of
q- 37 ø every 50-100 years, a great earthquake (mb7.0/Ms8.3)every
300-1000 years and a maximum magnitude event
(m•7.3/Ms8.8) every 700-2000 years. These estimates (and
those developedin this study) assumethat the historical earth-
quake listings [e.g., Nuttli and Brill, 1981] are free of extended
aftershocks of the 1811-1912 New Madrid sequence.They
56 ø
also assumethat New Madrid behavesin a time-predictable
ß manner, as has been demonstratedfor both interplate [Bufe et
al., 1977] and intraplate [Shimazaki and Nakata, 1980] fault
zones.Sincea prehistoricchronologyis not available for New
Madrid, we cannot verify such behavior; it remains the funda-
mental assumptionof this paper.
A Current Frequency-MagnitudeRelation
for New Madrid
Data Base. The historical seismicitylisting compiled by
Nuttli and Brill [1981] was combined with the instrumental
record of the Saint Louis University seismicnetwork [Stauder
et al., 1974-1983] to estimate mean recurrence rates for the
New Madrid seismic zone. Two different source zones were
usedin this study (Figure 1), an approach taken in responseto
the work of Wesnouskyet al. [1983]. On the basisof Quater-
nary fault slip rates in intraplate Japan, Wesnouskyet al.
[1983] suggestthat the observedlogarithmic distribution of
earthquake magnitude (equation (1)) is a consequenceof a
logarithmic distribution of individual fault dimensionsrather
than a logarithmic distribution of earthquakesizeson a single
fault. Thus it is not critical (or even advisable) to restrict the
sourcezone to only the most active portion of New Madrid,
(1)
even though it is probable that major earthquakescould only
occur there. The smaller source zone is about one fifth the size
of the larger zone yet contains60% of the historicaldata and
76% of the instrumental data. It is representedby the small
rectangleof Figure 1. Table 2 lists the area and number of
eventsfor each sourcezone. Arguments favoring selectionof
the larger sourcezone for New Madrid recurrencerates are
presentedin the final section.
A cumulative frequency-magnitudeplot of the combined
historical and instrumental data sets for both source zones is
shown in Figure 2; it was generatedfollowing the technique
outlined by Johnston[1981]. From the instrumentaldata set,
9 years of data were available, ranging from a completeness
thresholdmagnitude of 1.7 to a maximum of 5.0. These data
were combined with the historical listing, 158 years of data,
ranging from a thresholdmagnitudeof 3.6 to a maximum of
6.2. In order to combine the data sets,each set was normalized
to cumulative eventsper year.
The transition from instrumental to historical data was
made between magnitude 3.5 and 3.6. This assured a near
completenessof both data sets,as reportingof low magnitudes
by Nuttli and Brill [1981] is complete to m•3.6 [Johnston,
1981] and instrumental magnitudes above 3.5 for which 9
6740 JOHNSTONAND NAVA.' NEW MADRID PROBABILITIES

TABLE 2. New Madrid Frequency-Magnitude Parameters

T. )•ears
Method Increment a aa b ab (mb_>6.0) (m• _>7.0)

Large SourceZone

Least
squaresAM AM =0.5
-- 0.13.550
3.346 (+.037)
(+.070) 0.882
0.882 (+0.028)
(+0.030) 88
55t 70 420)
673T 550
Maximum likelihood* AM = 0.1 3.862 -.. 0.937 (.1_0.032) 58 498
[Bender, 1983] AM = 0.5 3.803 ... 0.909 (+0.032) 45 363
Small Source Zone

Least
squares AM=0.5
AM= 0.13.488
3.187 (-!-0.055)
0.896
(+0.095)
0.930(___0.046)
155•
(___0.043)
124) 1216
t 1100
140 1052
Maximum likelihood* AM = 0.1 3.899 --. 1.019 (___0.053) 164 1714
[Bender, 1983] AM -- 0.5 3.851 ... 1.001 (+0.050) 143 1432

Data setfor largesourcezone,157,700km2 (Figure1),956 events'1816-1983,magnitude1.7-6.2'data

setfor smallsourcezone29,960km2(Figure1),685 events;1816-1983,magnitude1.7-6.2.
*Included for comparison only--see text.

years may not be an adequate sampling time are excluded weight to the upper magnitude points that it is less suitable
from the instrumentalportion of the plot. than least squaresfor estimating the recurrenceintervals of
Methodology. Both linear regressionand maximum likeli- high-magnitude infrequent earthquakes. (See Shi and Bolt
hood techniques were used to determine the Gutenberg- [1982] for a discussion.)This is especiallytrue for large data
Richter constantsa and b for the "best fit" line through the setssuchas usedin this study. For large well-defineddata sets,
data. Each technique has advantagesand disadvantagesthat then, cumulativeleastsquaresprovidesa reliable estimationof
have been extensivelydiscussedin the literature [e.g., Wei- the data [cf. Welcheft, 1980].
chert, 1980; Bender, 1983]. As pointed out by Weichert, con- On the basis of the arguments of the foregoing paragraph,
ventional least squares (LS) is the maximum likelihood we choseto baseour derived frequency-magnitudeformula on
method for independentdata with a Gaussian error distri- least squaresregression.Table 2 containsthe constantsa and
bution. For cumulativefrequency-magnitude plots as usedin b of equation(1) and their uncertaintiescomputedby both LS
this studytheseconditionsare obviouslyviolated. and ML for magnitudeincrementsof 0.1 and 0.5 for both the
Conceptuallymaximum likelihood (ML) is the superioresti- large and small sourcezone.
mation technique;in practice,however,ML accordsso little Statistical uncertaintieswere computed as follows. For the

iooo iooo
[ [ ! I I I ! ! | -

(o) Lorge Source Zone - - (b) Sinoil Source Zone

••%c• 1816--1983 _•_

_•':m
• 1816--1983

• •o ; INST. / HIST. • ,o
• ß TRANSITION
ß INST.
/ HIST.
? TRANSITION

i i
u
z

....... 10

___

I 2 3 4 õ 6ø I 2
MAGNITUDE
(mb) MAGNITUDE
Fig. 2. Cumulativefrequency-magnitude data basefor determiningthe Gutenburg-Richterconstantsa and b. Instru-
mentaldata from 1974to 1983wereusedfor m• < 3.5; historicaldata for m• _>3.6.Opensymbolsdenotedata pointsnot
usedin computation.(a) Data basefrom largesourcezoneof Figure 1. (b) Data basefrom smallsourcezone.SeeTable 2
for computed recurrenceformulas from thesedata.
 JOHNSTONAND NAVA' NEW MADRID PROBABILITIES 6741

iooo
Large zone

log (Nc)= 3.43-0.88(rnt,) a,, • 0.060 at, - 0.030

tiI1--• Large
Source
Zone
(4a)
Small zone
ioo
-- _ _ 1816--1983
log (Nc)= 3.32-0.91(rnt,) 0.090 at, -• 0.045
(4b)
PROBABILITY ESTIMATES

A problem of increasingconcern in the central United

States is the likelihood of occurrence of the next large New
Madrid earthquake.(This is a separateand simpler question
o
-I.- than that of finding the total seismichazard which entails
i-
estimatingthe levelof groundmotiongeneratedby multiple
source zones and the chance of exceeding it at a given lo-
i
cation.) In this sectionwe will examine how well probabilities
of occurrence can be estimated for the New Madrid seismic
\\
zone.

The averagereturn period or recurrenceinterval, as derived

in the previous section,does not in and of itself supply suf-
ficient information to determine probability of occurrence.It
is also necessaryto know the frequencydistribution of recur-
rence intervals for a given magnitude or magnitude range.
That is, for a time span much greater than the estimated
recurrencetime, what is the distribution of actual repeat times
about the estimatedmean? Given the mean, the type of distri-
I 2 5 4 5 6 bution, the standard deviation of the observations, and the
MAGNITUDE(rnb) time of occurrenceof the last event, either cumulative prob-
Fig. 3. Illustration of the effect of varying AM, the magnitude
ability (the probability that an earthquakewould already have
increment.Magnitude incrementsof 0.1 (solid symbols)and 0.5 (open happened) or future (conditional) probability may be ascer-
symbols)are shown with least squaresline fit. tained.
For the New Madrid seismiczone, neither the type of distri-
standarderror of a, with n magnitudeintervals,xi = ith mag- bution 'of recurrence intervals nor their standard deviation
from the mean is known. Because of this lack of essential
nitude interval and yi = log N•,
information we can only proceed by assuming that New
Madrid behaves in certain ways. Our approach is to take a
aa= • [y•-(a- byi)]2/(n
- 2)
i=1
range of assumeddistributions and standard deviations and
examine the effect of theseassumptionson probabilisticesti-
mates. If the final estimatesdo not diverge too much, we can
{['E Xi2•
ß
i=1 i
Xi n
\i=l
Xi2 n have confidencethat the determined probabilities are fairly
stableor robust,i.e., insensitiveto the initial assumptions.This
(2) presumes,of course,that our assumptionsare comprehensive
enough to encompassthe actual behavior of the New Madrid
[e.g., Miller, 1982]. For b the standard error of b for a larger seismic zone.
number of eventsfrom $hi and Bolt [1982] is used:
Terminolooyand Definitions
at,= 2.30b2 (rni- r•)2/n(n- 1) (3) Before examining individual cases we define some fre-
i=1
quently usedterms and concepts.
Probability. The statement "x has the probability P(x)"
where r• is the mean magnitude corrected for grouped mag-
means that if an operation (read "earthquake occurrence")is
nitudes following Bender [1983]. Magnitude group size can
repeateda great number of times, it is practically certain that
significantlyaffect recurrenceestimates,as shown in Table 2
the relative frequencyof x is equal to P(x). Here relative fre-
and Figure 3.
Also includedin Table 2 are mean recurrenceintervals T• of
quencyis just the number(range)of actualoutcomesdivided
by the number (range)of possibleoutcomes.
70 and 140 years for rnt,> 6.0 and 550 and 1100 years for
Cumulativeprobability distribution. If X is a ra,ndomcon-
rnt,>_7.0. These are the valuesof T• to be used in the subse-
tinuous variable that is a property of eventsof a population,
quent probability analysisand are intermediateto valuesde-
then for any real number x the distribution of X, denoted
terminedby LS for AM = 0.1 and 0.5. ML recurrencevalues
are included in Table 2 for reference but are not used for
F(x), is defined as the number (range) of all eventswith X < x
divided by the total number (range) of events.Therefore F(x)
probabilityestimation.
alsogivesthe probability of choosingat random an event with
A frequency-magnituderelation for New Madrid derived
X <_ x. Thus
from the least squaresanalysisand from the intermediate re-
currence times of Table 2 is F(x) = P(X <_x)
6742 JOHNSTONAND NAVA: NEW MADRID PROBABILITIES

P(A r• B)
Pc = P(BIA)=
P(A)

where P(A chB) is the intersectionof A and B, that is, a subset

of samplespaceconsistingof all possibleoutcomescontained
in both A and B. In the context of earthquake recurrence,A
and B are events in the same magnitude range and [Pc =
Prob. next event (B) within (t, t + At)lPrecedingevent (A) at
tO].
A useful formulation of conditional probability is in terms
of cumulative probability or area beneath the curve of a prob-
ability density function. This is illustrated in Figure 4 for an
(b) arbitrary distribution F(t). Here Pc is just the ratio of two
areas' one area beneathf(t) betweent and t + At representing
the probability of occurrenceduring this time interval; the
- F(t):AF(t) other is the area beneathf(t) from t to + •, representingthe
probability that the event will eventuallyoccur given that at
time t it has not yet happened.Thus

- oo t t+At ,oo Pc =
[i f (z) d'r
]/[i© ] f (z) d'r (8)

_ AF(t)
• i- Fit) or from equations(5) and (7)

Fig. 4. (a) An arbitrary distribution function illustrating the con- r(t + at)- r(t) ar
Pc = = (9)
cept off(t), the probability densityfunction, and F(t), the cumulative 1 - F(t) 1 - F(t)
probability function. The variable t denotes time. Note that for nor-
malized or standarddistributionsthe total area betweenf(t) and the Frequently the computation of Pc is simplified by using
abscissais 1. (b) Formulation of conditionalprobabilityPc in termsof F'(x) = f(x) (from equation(6)) and expandingF(x) in an in-
the ratio of two areas beneath a standardizedprobability density crementalTaylor series'
function.
Ax 2
F(x) is the cumulative probability distribution (or distribution AF= r(x+ Ax)-F(x)= Axr'(x)
+ • f"(x)+... (10)
function) for an arbitrary distribution F and arbitrary variable
x. In this study we are interestedin variations with time t and
Thus, for example,for a Gaussianconditional probability a
first-orderapproximationis
assign specificcumulative distributions as follows: Gaussian
(or normal), G(t); lognormal, L(t); Weibull, W(t); Poisson,S(t) AG Atg(t)
Pc =•-•• (11)
(to avoid confusionwith cumulativeprobability P(t)). 1 - G(t) 1 - G(t)
Probability density. If x is a continuous variable, then
Care must be taken, however,to apply this approximation
there existsa functionf(x) suchthat
only for At <<t' otherwise,unreasonablylarge conditional
probabilitieswill be obtained(in extremecasesPc will exceed
r(x):
Ixf() (5) 1). Figure 5 illustratesthe differencebetweencomputingan
approximate Pc from equation (11) and an exact Pc from
where • is just the variable of integration; f(x) is called the equation (9).
probability density.It is just the derivative of the cumulative
distribution function F(x): 120 /

At :50yr /
F'(x) =f(x) (6) IOO Approx/ At=50yr(60T
r)
Figure 4a illustrates the concept of the cumulative distri- / Exac•x /
/• //////
8O
bution and the probability density functions.Note that since
F(x) = P(X < x), if events in the sample spaceare mutually
exclusiveand collectivelyexhaustive,it follows that • At:15yr . /

P(x)=
I_+••f
(x)
dx=l 40

=o

25 .50 75 I0 I 25 150 175 2 0

Specificprobability densitiesin this study are denoted as fol-

lows: Gaussian, g(t); lognormal, /(t); Weibull, w(t); Poisson, Fig. 5. Conditional probability by approximate and exact meth-
s(t). ods. The Gaussian distribution for an mb -6.0 quake is chosen for
Conditionalprobability. Let A and B be two events(or two thisillustration.The approximation
methodis the first-orderTaylor
setsof outcomes)in a samplespace.Conditional probability is expansion of thecumulativeprobability
(equations
(10)and(11));the
defined as the probability of event B occurring given that exactmethodistheratioof areas(equations(8)and(9)).Notethatthe
first-orderapproximationis a good representation of conditional
event A has already occurred. It is denoted by P, = P(BIA).
probability
onlyforsmall
Atandforelapsed
times
t less
thanT,,the
Thus P, isjust the ratio averagerepeat time.
 JOHNSTON AND NAVA: NEW MADRID PROBABILITIES 6743

IOO

NEWMADRID
(SmallSource
Zone)

Ili 1974
-1981:
530events
mb->
I'?
40- e -t•, T•:6.2
days

.5 I0 15 20 25 30 35 40 45 50
INTEROCCURRENCE TIME -
•days)

Fig. 6. Observed
interoccurrence
timesof theinstrumental
NewMadriddataset[Stauder
et al.,1974-1983].
A Poisson
distribution
functionin termsof interevent
times(equation(12))is shownfor comparison.

SpecificProbability Distributions Substitutionin equation(9) yieldsa conditionalprobability:

Four possibleprobability distributions were consideredas Sc(t,At) - 1 - e- At/T, (14)
representationsof the actual recurrenceinterval distribution of
New Madrid earthquakes for a given magnitude range. Each where At is the time interval under consideration for an earth-
distribution has seen previous use in seismichazard analysis. quake reoccurrence.Note that for a given At, Sc is constant
In the absenceof an actual New Madrid chronology, we feel and does not depend on t, the elapsed time since the last
that this approach has a high likelihood of bracketing the event.
actual behavior of the New Madrid seismic zone. Gaussian. The Gaussian (or normal) probability density
Poisson. The discretePoissondistribution has the proper- function is given by
ty that the probability of occurrenceof an event is exactly the
samefor a given time interval anywherealong the time axis.
(15)
This timeostationarityproperty is usually expressedin a distri- •/(t)
- o.(2•r)•/2
exp--• o'
bution function for the number of eventsper unit time. More
useful for our purposesis the distribution of interevent times wherea is the standarddeviationand T• is the meanrepeator
which may be shown to follow a negative exponential distri- recurrencetime. The cumulativeprobability is
bution for a Poissonprocess[Lo/nnitz, 1974]:
G(T _<t) = g(r) dr (16)
=•1e-t/Ti
s(t) (12)
This integralcannot be evaluatedby standardtechniques,but
where T• is the average interevent time. Figure 6 shows the extensivetablesof valuesare availablewhen G(T < t) is trans-
distribution of instrumental New Madrid seismicity and the formedinto a standardizednormal distributionby using
Poisson probability density function for T• = 6.2 days and
/no_>1.7. t-T•
z- (17)
Poisson statisticshave been used extensivelyto represent
time sequencesof earthquakes. In many cases,seismiczones
From equations(9) and (16) the Gaussianconditional prob-
seem to emulate closely a Poisson process (e.g., Figure 6). ability is
However, this is a different property from the one of interest
here, which is the distribution of occurrence times about an G(z + Az)- G(z)
averagerecurrencetime for a givenmagnituderange.For this, Gc(t,At)--} Gc(z,Az) = (18)
1 - G(z)
Poisson probabilities are not an adequate representationbe-
causethe independenceof Poissoneventsresultsin a constant Values for G(z + Az) and G(z) were taken from National
conditional probability rather than one increasing in time Bureauof Standards[1953] for this report.
sincethe last event as required by the time-predictablemodel The Gaussian distribution has been used recently by a
assumed for New Madrid. number of authors to estimate conditional probabilities of
Even though Poissonprobabilities are not appropriate for seismiczonesin Japan and the United States [Wesnouskyet
this study, their extensiveuse in the seismichazard literature al., 1984; Sykesand Nishenko,1984; Jacob, 1984]. In these
make them useful as a referencelevel when time-dependent studies,either long historical chronologies,geologicalslip
conditional probability is discussed. rates,or, paleoseismicdata were available to placeconstraints
The cumulative Poisson probability for a time interval on either the estimatesof mean recurrencetime or the appro-
T _<t and for an averagerecurrencetime T• is priate distribution to apply. Gaussianconditional probabil-
ities will be an important componentof this paper, but we
S(T < t)= s(r)dr= 1-e -t/T' (13) emphasizethat no independentempirical or theoreticalargu-
ments exist that favor their use for New Madrid.
6744 JOHNSTONAND NAVA: NEW MADRID PROBABILITIES

TABLE 3. Weibull Constants

where F is the gamma function. The v is often referred to as
Average
the shape parameter and increasesas a decreases.The 2 is
Recurrence Standard 2 v exponentiallyrelated to the mean rate of failure and increases
Interval Deviation Rate Shape as T• decreases.Table 3 liststhe Weibull parameterscomputed
T,, years a(% of T,) Parameter Parameter References for this study from equations(21) and (22) as well as some
found in the literature.
35 24 years 0.0061 1.41 Kiremidjianand
(69%) Anaonos[1984] Equation (20) may be directly integrated to obtain the cu-
87 34 years 3.52x 10-6 2.70 Ha•iwara[1974] mulative Weibull probability:
(39%)
70* 33% 5.69x 10-7 3.30 thisstudy W(T _<t) = w(z)dr = l - e- x" (23)
70* 50% 1.03x 10-'• 2.10 thisstudy
140' 33% 5.78x 10-8 3.30 thisstudy
140' 50% 2.41x 10- 5 2.10 thisstudy which yieldsa conditionalWeibull probability of
550•' 33% 6.32x 10-•ø 3.30 thisstudy
550•' 50% 1.36x 10-6 2.10 thisstudy exp [-2t v] -exp [-2(t + At)v]
1100'{' 33% 6.42x 10-• 3.30 thisstudy W•(t, At)= (24)
exp [- 2t v]
1100'{' 50% 3.18x 10-7 2.10 thisstudy
We note that the Weibull hazard rate, defined as
*The mb _>6.0.
•'The mb_>7.0.
dW(t)/dt- 2vtv-x is often usedas a conditionalprobability
[e.g., Kiremidjan and Anagnos, 1984; Hagiwara, 1974]. This
expressionis just a first-order Taylor expansionand can lead
Lognormal. The'lognormal distribution hasfound frequent to significantoverestimatesof W•(t, At) in the samemanner as
application in the earth sciences[e.g., Till, 1974] and in at
shownfor the Gaussiandistribution (Figure 5).
least one case has been applied to earthquake recurrence
times. Jacob [1984] compared the fit of the interoccurrence
intervals of large (M, _>7.8) Aleutian earthquakesto both Cumulative and Conditional Probabilities
Gaussianand lognormal distributionsand consideredthe log- for New Madrid
normal to be the more appropriate representation.It is includ- In this sectionwe presentprobabilityestimatesfor large
ed in this study as a potentially valid distribution for New (mb>_6.0) and great (mb>_7.0/M, >_8.3) New Madrid earth-
Madrid recurrence times.
quakesbasedon the informationpresentedin the preceding
The lognormal distribution can be generated from the sections.
Gaussian(or vice versa) by the variable transform t = In (t), Cumulativeand conditionalprobabilityestimatesrequire
whereIn is the natural (basee) logarithm.A lognormaldensity choiceof a probabilitydistribution,specificationof the distri-
function l(z) is obtained from the standardizednormal distri- bution mean, and specificationof the standard deviation of
bution usingthe following relations: observations about the mean. Conditionalprobabilitiesalso
requirethe lengthof the futuretime interval At. Figure 7 is a
z= •* In(t)-T•*a*=
[ln
(.a2
+2T•2•]
TM •' 'J3
parameter tree depiction of the various parameter combi-
nations that were chosenfor analysisand illustration. The
a'2 (19)
choiceswere dictated by a variety of factors as discussed
T•* = In (T3- • below.
2
Magnitude. We restrictour data presentationto two mag-
where T•* and a* are the mean and standarddeviation of the nitudes:an m,6.0(M,6.3) eventwhichcouldbe locallydestruc-
lognormal distribution. tive over severalstate countiesand m,7.0 (M,8.3) which we
Weibull. The Weibull distributionwas originallydevel- take as a great 1812-typeNew Madrid event,destructiveover
oped by Weibull [1951] on a purely empiricalbasisfor appli-
severalstates.(We follow Nuttli's [1983] study of midplate
cation to instancesof failure of individual componentsof large
earthquakesfor all mdM• magnitudeconversions.) A maxi-
systems.Hagiwara [1974] and Rikitake [1975] applied thismum magnitudequake (m,7.3/M, ,,, 8.8 [Nuttli, 1983]) has
distribution to data on crustal strain precedinglarge earth-
suchlow cumulativeand conditionalprobabilitiesas to be of
quakes. If the strain rate is approximately constant (as re- little interestfor this study.
quired by the time-predictablemodel), a Weibull distribution Averagerecurrence interval. In the absenceof any geologi-
of "ultimate strain" will allow estimatesof probability of oc- cal or paleoseismological informationthis parameteris taken
currence. Since this Japanese work, Weibull statistics have directlyfrom the frequency-magnitude analysisof the first part
been increasingly applied in seismic hazard research [e.g., of this paper. As is evidentfrom Table 2, recurrenceestimates
Brillinger, 1982; Kiremidjian and Ana•lnos, 1984; Nishenko, for a givenmagnituderangehave a large range.Dimensionsof
1985]. the sourcearea and the line-fitting techniqueare the two prin-
The Weibull probability densityfunctionis givenby cipal sourcesof this variability. Differencesin the size of the
chosensourcezone causedifferencesof roughly 100% in the
w(t)= 2vtTMxe- a,, (20)
estimates of mean recurrence times, and different line-fitting
where 2 and v are constantsthat are related to T. the mean methodsyield variationsof the order of 50%.
time to failure. and to a. the standard deviation. as follows Such a large range of estimatedrepeat times must be con-
[Hagiwara. 1974]: sideredin any realisticprobabilisticanalysis.Our approach
was to take a value from the low and from the high end of the
rangeof estimatesof T• for usein the probabilityanalysisand
T•= ©
tw(t)
dt=k-x/T
(v+ 1) (21)therebybrackettheactualbehaviorof NewMadrid.Extreme
pointswere not selectedbecausethis would smearthe results
beyondusefulness and becauseextremesare accountedfor to
a degreeby selectinga rangeof standarddeviationsabout the
 JOHNSTON AND NAVA.' NEW MADRID PROBABILITIES 6745

section) and is exhibited by virtually all seismic zones that

have been identified as behaving in a cyclic manner. Table 4
presentsa representative(but not exhaustive)listing of repeat
times and standard deviations that have been reported in the
literature for various seismic zones worldwide.
Information such as that in Table 4 is not available for New
Madrid; therefore standard deviation must be an assumed
parameter. Using Table 4 as a guide, the standard deviation a
is allowed to vary from one third (33%) to two thirds (67%) of
T• (Figure 7). For o' in excessof 0.5T• the very conceptof the
time-predictable seismicitymodel loses much of its meaning.
The observed variability of the repeat times of magnitude 5
and 6 earthquakes in the historical record [Nuttli and Brill,
1981] suggeststhat o' should not be smaller than one third of

Type of distribution. Type of distribution is another as-

Range
ofPc sumed parameter for New Madrid. As previously discussed,
we selected four commonly used distributions for analysis.
One of them, the Poisson, is included only for reference,as it
is not applicable to a time-predictable model.
For New Madrid earthquakesof body wave magnitude 6.0
and above, the elapsed time since the last event(s)is still less
than the estimated repeat time (with the exception of T• - 70
years). The analysis of this section will show that for such a
time frame, choice of probability distribution makes little dif-
ferencein computed probabilities.Sykesand Nishenko [-1984]
reach a similar conclusionfor severalCalifornia fault systems.
Time interval At. The choice of a time interval for com-
puting conditional probabilities is completely arbitrary since
the exact form (equation (9)) of Pc rather than the first-order
Taylor series approximation (equation (11)) is employed in
this study. We selected a At of 15 years to coincide with a
Fig. 7. Logic tree of the five parameterswhich are allowedto probability of occurrenceby the turn of the century and a At
vary to obtaina rangeof estimates of conditionalprobabilityPc.The
branchesare labeled as follows:(1) is for magnitudesof mb6.0(Ms6.3) of 50 years as representativeof the probability of occurrence
and mb7.0(MsS.3), (2) averagerecurrence interval(in years),(3) stan- during a lifetime.
dard deviationas a percentage of the averagerecurrence interval,(4) Cumulativeprobability. It is of greaterinterestto know the
the assumedprobabilitydistribution,and (5) time incrementin years probability of a large earthquake happening during some
over which conditionalprobabilitiesapply.
future time interval than to know the probability that it would
have already happened by now (the present). For this reason
estimatedmean (seenext section).Thus in Figure 7 a range of we emphasize conditional rather than cumulative probabil-
70-140 yearsis usedfor T• (r%6.0)and 550-1100 yearsfor T• ities. We do include, however,the general behavior patterns of
(r%7.0)(seeTable 2). cumulativeprobability for the four distributionsconsideredin
Standard deviation. Frequency-magnitudeanalysis yields this study.Theseare illustratedin Figure 8 and summarizedin
an estimatedrecurrencetime T• but no estimate of the vari- Table 5.
ation of T• as the seismiczone proceedsthoroughmany seis- In Figure 8a, cumulativeprobabilitiessince 1895 are shown
micity cycles.This variability is physicallyreal (as opposedto for an m• > 6.0 quake with a mean recurrenceinterval from
the variability inducedby methodologyas discussed in the last the extremeupper end of its range (T• - 140 years).Gaussian

TABLE 4. Worldwide Mean Recurrence Times and Standard Deviations

Mean Recurrence Standard Deviation

Interval
Location Reference T,, years Years Percent of

Southern San Andreas Sieh [1978] 164 76 46

fault, nine events
Southern California Sykesand Nishenko [1984]
Parkfield 22 7 32
Pallet Creek (12 events) 145 61-73 42-44
Imperial fault 32 10 32
Aleutian Arc, Alaska, Jacob [1984] 73 32 44
28 events
Oaxaca, Mexico McNally and Minister [1981] 35 24 69
Nankai-Tokai, Japan Rikitake [1977] 170 68.9 41
Hokkaido-Kuriles Arc Rikitake [ 1977] 85.3 24.6 29
Northern South America Rikitake [ 1977] 46.3 30 65
Southern South America Rikitake [1977] 100 22.5 23
South Kanto, Japan Hagiwara [1974] 87 34 39
 6746 JOHNSTON
ANDNAVA: NEW MADRIDPROBABILITIES

I00

9O

8O
Mb6.0(Ms63) Log

• 6o

Polsson i
.>- •o

20

I0

? 2o 4o $o •o •oo •ao •o •o •o aoo aao

/895

Ioo
(b)

90

8o

TR:550yr

Gaussian
TR=1100yr

._> 4O I
Mb7.0/Ms
8.3
/995

E•o

2O

2O0 40O 600 8OO I000 1200 1400 1600 18002000 2200 24'00;
Elopsed Time (yrs)
Fig. 8. (a) Cumulativeprobabilitiesfor the four distributionsconsideredin this studyfor the representative
caseof a
quakern•> 6.0, meanrecurrence time T• of 140 years,to = 1895,and standarddeviationof 0.33T•.The Gaussianand
Weibulldistributions
are nearlyidenticalin thiscase.(b) Illustrationof the variationof cumulative probabilities
with T,
and a for a quakern•> 7.0, to = 1812,for the Poissonand Gaussiandistributions. The Weibulland lognormaldistri-
butionswould vary a relativelyminor amount from the Gaussian.

and Weibull valuesare within 1% and vary from the lognor-
mal by a maximumof 5%. Poissonvaluesare significantly
higher than the other distributionsfor t < T• and lower for
t > T•. This behaviorholds for all choicesof T• and a and
carriesover into the conditionalprobabilityestimates.
Figure 8b illustratesthe dependenceof cumulativeprobabil-
ity on T• and a for an rnb> 7.0 earthquakefor which the last
event was in 1812. For the range of recurrenceintervals and
standarddeviationconsidered in this study,cumulativeprob-
abilitycanvary up to ,-•60% with T• and ,-•10% with •. Only
the Gaussian and Poisson cases are shown since Weibull and
lognormaldifferences
from Gaussianare < 10% (Poissonis
shownfor referenceonly).Thus the estimateof T• has much
the greatereffecton calculatedcumulative(and conditional)
probability;choiceof a and choiceof distributionprovidea
modulation of only about 10%.
Table 5 summarizescumulative probabilities for New
Madrid for the year 1985.For the largesourcearea of Figure
1 (T• = 70 and 550 years)the probabilitythat an rnb> 6.0
quake would have occurredbetween1895 (the last rn•> 6.0
event, Charleston, Missouri) and 1985 has reached 72-83%.
Similarly for rn•> 7.0 since 1812, P(t = 1985) is only 0.3-
8.5%. If recurrence intervals derived for the small source zone
are considered
(T• = 140 and 1100years),the computedvalues
 JOHNSTON AND NAVA' NEW MADRID PROBABILITIES 6747

TABLE 5. Summary of Cumulative Probabilities' 1985

G½(t= 1985) L½(t= 1985) W•(t = 1985)

(Gaussian),% (Lognormal), % (Weibull), %
At Tr(ff1 = 0.33T,; a2 = 0.50T,), Pc(t = 1985) Sc(t= 1985)
(years) years (ExtremeRange),% (Poisson), %

rnb _• 6.0 (Ms6.3)

1895 70 _+(23-35) 80 72 83 78 81 73 72
11-83
(t -- t o - 90 years) 140 +_(47-70) 14.2 23.8 11.5 24.2 15.0 26.4 47

r% _• 7.0 (Ms8.3)
1812 550 _+(183-275) 2.0 8.5 0.3 1.3 1.5 6.6 27
<< 1.0-8.5
(t -- to - 173 years) 1100 _+(367-550) 0.6 4.6 << 1.0 0.011 0.16 1.6 15

fall to 11.5-26.4% and << 1.0- 4.6%. In all but the case for
T• = 70 years and the elapsedtime t = 90 years, Poissonsta-
tisticssignificantlyoverestimatethe likelihood of occurrence.
Conditionalprobability. Conditional probabilities for large
New Madrid earthquakes for the next 15 and the next 50
years are presentedin this section.Choice of values for the
parameterswhich affect the calculations(type of distribution,
mean recurrencetime, and standard deviation) have been pre-
viouslydiscussedand are summarizedin Figure 7.
The data are presentedin Figures 9a, 9b, 9c, and 9d, each
depictingconditional probabilitiesPc for Gaussian,Weibull,
lognormal, and Poisson distributions.Figures 9a and 9b are
for an mb> 6.0 earthquake with T• equal to 70 or 140 years;
Figures 9c and 9d are for an rnb_>7.0 event with T• equal to
550 or 1100 years.The plots have been carried out to elapsed
time double or triple T• so that asymptoticbehavior of Pc at
large t would be shown. Three referencetimes, to, T, and
t = 1985, are shown on each graph. Shaded regions indicate
the rangeof Pc for a givenAt as the standarddeviationvaries
from one third to one half of T•.
There are severalimportant featuresin Figure 9 to which
we call attention.
1. In all casesthe present(year 1985) is lessthan 1.5T•. In
this time range the choice of probability distribution makes
little differencein Pc. This is particularly true for the Weibull
and Gaussiancases;lognormal yields the lowest Pc for times
closeto to and risesmost steeplyas t approachesT•.
2. Allowing a to range from 0.33T• to 0.50T• has a greater
effect on conditional than on cumulative probability but only
for t > T, times generally beyond our concern in this study.
For times less than roughly one-half T• an increasein a in-
creasesPc becausearea beneath the probability density func-
tion is shifted from the distribution mean (T•) to the distri-
bution tails. However, for t > T, Pc for a = 0.50T• is always
lessthan Pc for a = 0.33T,
3. For elapsed time t >• 2T• the lognormal distribution
yieldssignificantlylower estimatesof Pc than either the Wei-
bull or Gaussian. This case appears to apply to the con-
ditional probabilities for some of the Aleutian arc segments
computedby Jacob[1984].
4. The constant Poisson conditional probabilities will
yield estimatesthat are consistentwith the other distributions
only over a very restrictedrange of elapsedtimes and should
not be usedfor conditional probability estimationfor a time-
predictable model of seismicity.In contrast, Poisson cumula-
tive probabilities are consistentlyhigh for t < T• and may be
used to provide a conservative(i.e., upper bound) estimate of
seismic hazard.
The 15- and 50-year conditional probability estimatesfor
the year 1985 are compiled in Table 6. The likelihood of a
great (r% > 7.0/Ms > 8.3) New Madrid earthquake in the next
15 to 50 yearsis uniformlylow, reachinga maximumof 4.7%
for a Weibull distribution with T• = 550 years, a = 0.5T,
At = 50 years.Lognormal estimatesare very low, often much
less than 1%.
Pc estimatesfor an m• > 6.0 eventexhibit a wide variation,
rangingfrom 9 to 67% for At = 15 yearsand from 34 to 99%
for At -- 50 years. About 10% of this variability arisesfrom
type of distribution (Poissonexcluded),10-25% from choice
of a, and from 25 to 50% from choice of T• = 70 or 140 years.
In the "restrictedrange"column of Table 6 we give a preferred
probability range taken from the low, T• = 70 yearsrecurrence
time (the large sourcearea of Figure 1). Argumentssupporting
this choiceare presentedin the concludingsection.
DISCUSSION AND CONCLUSIONS
The probability estimatesof this report rely on the assump-
tion that the New Madrid seismiczone generatesmajor earth-
quakes in a repeated fashion. It is further assumedthat the
occurrenceof large eventsis periodic rather than episodic,a
distinction we make on the basis of the standard deviation of
repeattimesabout the mean recurrenceinterval.If a is greater
than 50% of the mean recurrence time, we term the activity
episodicand our probability estimatesdo not apply. If a is
less than 50%, the calculated probabilities reported here
should be reasonableestimatesof the likelihood of future large
New Madrid earthquakes.
The 15- to 50-year conditional probabilitieslisted in Table
6 have a very wide range, e.g., from 9 to 67% for mo> 6.0,
At = 15 years,and from 34 to 99% for At - 50 years.Most of
the variability arisesfrom the choice of sourcezone bound-
aries for the seismicitydata set. We favor the choice of the
larger sourcezone (which leads to shorter recurrencetimes
and, consequently,higherprobabilities)and offer the following
argumentin supportof this selection.
The large sourcezone. According to Nuttli's most recent
estimates the three great New Madrid earthquakes of the
winter of 1811-1912 had surfacewave magnitudesof 8.5, 8.4,
and 8.8 [Nuttli, 1983]. Even neglectingthe extensiveafter-
shock sequences,these great eventsrepresenta strain energy
releaseof • 1.6 x 1025ergs.The questionwe pose is, what
crustalvolumeis requiredto storethis strain energyas elastic
potential energy?
If we considerthe simplestcasewhere strain is represented
by a singlecomponentof shear½), then strain energyper unit
volume (e) is
e = «#82
where# is the rigiditymodulus.Taking # = 5 x 10TMdyn
cm-2 and using Rikitake's[1981] most recent estimateof
ultimatecrustalstrain,ema•
= 4.4(+__
1.7)X 10-5, weobtain
e = 182-930ergs/cm
3
6748 JOHNSTON
ANDNAVA'NEWMADRIDPROBABILITIES

Mb6.0/ Ms6.5
TR- 70 years Last Occurrence' 1895

:::i:::::::::::::::::::::::::: .... .......?'.:.:;:7.:."..'..::'..'.

90 901 •--•/.:..:
i¾:.•:.:: ' '',,,,;>•-
"•'" /' ................ "' :• :••••••••:::.::•.":'

t •/' ' ==================================

•"::•:•:•:
'•::'
' '"":":'"::'
•.•/•?:::::::::::•:.•..•
[ • ...:.::.:?.::.??:::..:.:.::::".""
, ::ii:::.'• ..:..'..'..?:'..'..'.:..'.::'..7..?::..?
.....
eo ..:..
(.?;.?:'
ß
'ø
t ,'T. ..:{..:.,?..:::..:...
•o
ß , :.::.;.:::.:':...'::.
.'-"::'?.:..•...'::?•::'
.:..'i?.F.%'Y??.'h?:•"'

2 20

io io

1895 1895

ioo

Poisson

0 eo
At: 50yr
,

At: 15yr

ß , ß

? 20 40 60 E•C) tOO t20 140 IGO • ZOO2• 240

1895

Elapsed Time since Las• EvenI (yrs)

Fig. 9a. Meanrecurrence
interval,T•= 70years.

Fig.9. Computed
conditional
probabilities
forthevarious
parameters
shown
intheparameter tree
ofFigure
7.T•and
thecurrent
date
(1985)
areindicated
attheappropriate
elapsed
times
from1895,
thelastrnb> 6.0event
and1812,
thelast
r%> 7.0event.
Poisson
conditional
probabilities
areshown
forreference
only.

That is, typicalcrustalrockcannotstoremorethan 182-930 havearguedthat the crustcontains a ductileshearzoneat

ergs/cm withoutbrittlefailure.The 1811-1912 depths
3 elastically of •20 to •40 kmin whichductile deformation(of
events would haverequired a crustal
volumeof1.7-8.8x 1022 the orderof millimetersperyear)strainsthe overlying brittle
½m 3 to storetheenergyreleased asseismic (Uniform crust
radiation. (e.g.,
M.D. Zoback etal.,unpublishedmanuscript,1985).
strainenergy density
isassumed andwouldrep- Wetake25kmasa (possibly
forsimplicity high)estimate ofthethicknessof
resenta lowerboundon the requiredcrustalvolume.) the elasticcrust.This yields a crustalarea of 680,000 to
What portionof the crustis elastic? CentralU.S. earth- 3,500,000km2 requiredfor strainenergystorage.
quakes occurto depths of 20-25km.Recently,someauthors The areasof thelargeandsmallsourcezonesof Figure1
 JOHNSTON
AND Nnv^: NEW MADRID PROBABILITIES 6749

Mb6.0/ Ms6.3
TR- 140years Last Occurrence' 1895

K)O

Lognormal
(•: 33T. $o
to
/•85
• ==========================
:::::::::::::::::::::::::::::::::::::::::::::::::::::
.:..:.:.:.:f:!if:ii•::?',•;';':•'b'•;;'ii•i•i•!•ii•f:i
60
Poisson

'Jo j c7
,soT. 40

•o
:/.../ I• ......:.:.:.::::::::i:::!.i!!!!:!:ii!iiii:i!!:i:!:i:!:i:i
t--1.5
yr::::::::::::
:':':::::::
::.::i:::::::
•t: 50yr

? •
.•!4!:"'•'
40
! "'"'
•0
........:(•,
80
.5
øT,WO •ZO •40 •EO 18C) ZOO 220 240
At: 1,5yr

Elapsed Time since Last Event (yrs)

Fig. 9b. Mean recurrenceinterval, T, = 140 years.

are 157,700km2 and 29,960km2, respectively.

The required strainis 138,000km2 (for 25 km depth),still of the orderof the
crustal area for strain energyaccumulationfor great New dimensionsof the larger sourcezone.
Madrid quakesfar exceedsthe small sourcezone: It is for this By making this order-of-magnitude calculation for the accu-
reasonthat we favor the largerzone.Increasingthe depth of mulation volume of prior strain, we are not proposing that a
the ductile-brittletransitionto 40 km or assumingvaluesfor great earthquake could occur anywhere in the large source
/4 the rigidity modulus,above the already high value taken zone of Figure 1. We believe that such an event could only
would not changetheseconclusions. If an older,higheresti- occur on the principal fault segmentswithin the small source
mate of ultimate strain is taken, Emax •.• 1.5 x 10-4 [Tsuboi, zone. However, the strain storage limitations of the brittle
1956], then the requiredstrain energydensityincreasesto crust require the participation of crustal rock significantly
5625ergs/cm
3,andtherequired
crustalareato storetheprior beyondthe limits of the small sourcezone.This would, of
6750 JOHNSTONAND N^V^: N•W MADRID PROBABILITIES

Mb7.O/ Ms8.3
TR- ,.550years Las• Occurrence' 1812

•oo

90

Weibull
7O
,7'
/-""•"''
.'•.'.••.'.-'•-'.'
'"••,
...... •
::::::::::::::::::::::::::::::::::::::::::
.-
60

::::::::::::::::::::::::::::::::::::::::::::::
"':••'••.
''"'
......
'•.::.:.::::::i::ii:::
• ',:.?,
•"',.•:._'•-.................'..-...................
====================================
•,::•
:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.'
.,:.0::•?N-.'
'......'....'..'.'..'.'..'.'-.'.'..'....'..':..'.'-.'.'..'.'..'.).
T,R :::::::::::::::::::::::::::::::
:::::::::::::::::::::::::::::::::::::::::
:::::::::::::::::::::::::::: ..'.'.?.'.?.'.?:v:v:v:v.?.'.?:v.'v.'v.'•..?.'.?.'v.'v.'..v'::':.•

, I•ii:.•'•
'••••
'"
"'•"
'"'"'
'/" ''"
"'"
'"'"'
'"" !ii:.•l;i
!?..•:•'i:'-"
'".....'....-':••
"'"'"'
:.......•:•:i
""".-"
'"'"
20
I :. ß ..?:??.?:'.:':??.:':'.:'::,:'.:.:'.??.:':'-):'.:':?.v'.v":'"'"'
• ......•:!;i:i:!:'::;:'::::':'::':':'".':::'::":""
....

o 1•,203400r• 8C0I000
I• 1400
• • -•,• -"•.2f•
24•

Log normal
Poisson

At: 50yr
At: 15 yr

Elapsed Time since Last Event (yrs)

Fig. 9c. Meanrecurrence
interval,T, -- 550years.

necessity,encompassthe fault population of the large source dates 1812 and 1985 are used to bracket these events, an
zone; therefore we take the large zone as the proper one for averagerecur.rence of 58 (_ 30)yearsis obtained.Thesesparse
recurrence estimation. data are certainlymore compatiblewith the large sourcezone
Are the mean recurrence times of large events estimated recurrence estimates than with those of the small zone.
from the large sourcezone compatiblewith other lines of Another order-of-magnitudecheck can be made from the
evidence?From Table 2, an mr,>_6.0 event is expectedevery displacementvaluesthat Nuttli [1983] estimatesfor the great
55-88 yearson average(comparedto 125-155 yearsfor the New Madrid quakes.Accordingto Nuttli, sucheventshave
smallerzone).Since1812 two mr,>_6.0 eventshave occurred: displacements of 5-8 m. If a representative
sourcedimensionis
in 1843 near Marked Tree, Arkansas, and in 1895 near -• 100 km, this represents
a strainof 5-8 x 10-•, whichis
Charleston,Missouri. The intereventtime is 52 years; if the close to the ultimate strain estimate of Rikitake. A repeat time
 JOHNSTON
AND
NAVA:
NEWMADRID
PROBABILITIES
6751

Mb7.o/Ms8.5
TR- I,I00 years Last
Occurrence'
1812

90

Weibull

•0 I . . '...'.
'..:'.
.:'.'...'..'.::
....::,..,'.:...
'....:....:,
'.•7.:
' /98.5
'"1 ' I •,,"'""--"
'••••.:..
-•'
''''•'"
"'"'"''
'"'"'
'ß ß ' '"' :::.t:.:
,oI , I ..•:t:.?ttt.:.:??:?'.':.
. :.:..
t•:.•.•'tt::::t"---'
'"'
"•:-
••0 •of_
_•00
/•2
400
r•o
eoo
•r,,'•,
.••400
--•001800
• •zoo
•'•'.,
2000
2200 I.......
I "•••'•.. . ... ..•i'""':;:•":':•":
....:.•.::::::::::':':':':':'"'"
......'.'
'..'.::'.'¾'.
1•22004006008(10
•00•200•400• t800 •0022•240)

0 Lognormal ,o

Poisson

/, I ...•.•.•.••_..•_:.:.•.
•. ..... •o

t0 r--- •0
2400
ElapsedTimesinceLast Event
(yrs)
Fig.9d.Mean
recurrence
interval,
T•= 1100
years.

of550years
requires
astrain
rate
ofg_•1-1.5
x10-7
yr-•,
a
highbut
notunprecedented
value.
Harada[1978]
foundfor energy
ofgreat
NewMadridearthquakes.
Therecurrence
the
south
Kanto
district
ofintraplate
Japan of1x 10-7 rates
arate arecompatible
with
theoccurrence
ofrn
0_>
6.0events
in
yr-xover
a 70-year
period.
Normal
strain
rates
foractivethe
historical
record,
with
displacement
estimates
forgreat
interplate
zones
areatleast
nitudelarger. afactor
of5toanorder events,
ofmag- withestimates
ofultimate
crustal
strain,
andwith
in-
traplate
strain
rates
inJapan.
The
location
ofalower
mag-
Ouroverall
conclusion
can
bestated
asfollows. nitude
Recurrence 6eventwouldnotberestricted
tothesmall
source
rates
derived
fromthe
large
source
zone
ofFigure
1are area,
pre-on whereas
agreat
earthquake
isconsidered
possible
only
ferred
tothose
ofthesmall
source
zone
because large theprincipal
ofthe fault
segments
oftheNewMadrid
zone.
volume
ofcrustal
rock
required
tostore
theelastic
potentialRestricted
probability
range.
InTable
6wehave
included
a
restricted
range
ofconditional
probabilities.
thesevalues
are
6752 JOHNSTON
AND
NAV•:,
NEWMADRID
PROBABILITIES

................. ,-......... •,,.•

TABLE
,
6.,,, Summary
, ,
ofConditional
,-,,, ? • ,, ........
Probabilities:
1985
ac(t, At) Lc(t, At) W•(t,At)
T,(•x -0.33T,; (Gaussian),% (Lognormal),% (Weibull), %
At, a2 = 0.50T•), Pc(t,At) Pc(t,At) $•(t, At)
years years a• a2 a• a2 a• a2 (Extreme Range),% (RestrictedRange),* % (Poisson),%

mb >_6.0 (M,6.3)

15 140
704-+(47-70)
(23-35)66
9.9 44
9.3 55
13.5 38
14.8 67
10.2 39
11.0 t 9-67 40-63 19.3
10.2

50 140
704-__
(23-35)42
(47-70) 99 92
34 94
51 80 42
46 99 3787t 34-99 86-97 51.0
30.0
mb>_7.0 (M,8.3)

15 1100
550 +(367-550)
+ (183-275)0.440.28
0.07 0.97<<0.05
1.0 0.750.05
0.01 0.5 0.30
1.3t <<
1.0-1.3 0.3-1.0 2.7
1.4

50 1100
550 4-(367-550)
+ (183-275) 1.8 3.5
0.27 1.0 0.4 0.07
<<1.0 3.4 0.2
2.0 4.7
1.1 I <<
1.0-4.7 2.7-4.0 8.7
4.4

*See text.

based on the arguments of the previous discussionin favor of Acknowledgments.

An earlyversionof this work waspresented
at
largesourcezonerecurrence
intervals,i.e., T, = 70 yearsfor theNewMadridSeismic
ZoneSymposium,
April27-28,1984,Cape
mb>--
6.0andT,= 550yearsforrnb>
--
7.0.Ourrestricted
range Girardeau, Missouri•sponsored bytheU.S.Geological Survey and
the Missouri Academyof Sciences.We wish to acknowledgethe De-
estimates aresimplyan averageof therangeof values ob- :•'ji•mentofEarth andAtmospheric SciencesatSaintLouis Univer-
tainedwiththeGaussian, lognormal, andWeibulldistri- '•;•.and,'in
'particular,
O.Nuttlifortheexcellent instrumentaland
butions andaregivenin Table7 asoneof themainCo n- •"oxical seismicity
databases used in thisstudy. Wethank S.T.
clusions ofthisstudy.
Theestimates foranmb6.6/Ms7.6event .f•iss en,S.P.Nishenko, and J.E.Zollweg forvaluable
discussion
were
derived
using
thesame
methodology
asforthemb6.(Y
•aia61.
•'•.a•
•"eview
,•',•imum and B.Bender
likelihood foracopy
estimation ofher
Computer
of grouped progra;Va•:'
magnitude for
•data.This
m•7.0
events
andareinciuded
here
forreference
bemuse
ofl•'l• •'•dybefiefited
consid•ably
from
themamascript
preparation
ofR.
suggestion
thatsufficient
strainhasaccumulated
since
1812to,;! Pryor.
Research
support
wasprovided
by•t-he
Nuclear
Regulatory
cause
anMs7.6NewMadridevent[Nuttli,1983]. •:' COmmission
under
contract
NRC-04-78-20•and
bytheState
often-
These probability estimateswould be much lower if small
sourcezone statisticswere appropriate: for m• > 6.0, ,-•10%
REFERENCES •..;i.
by 2000 A.D., and ~ 39-45% by 2035 A.D.; for m• ___ 7.0,
<<1% by 2000 A.D. and 0.2-1.0% by 2035 A.D. Algetmissen,S. T., The seismicrisk map of the United States:Devel-
The uncertaintiesinvolved in deriving the above estimates opment,use,and plansfor future refinement,in Conference on Seis-
mic Risk Assessment for Buildinq Standards(March 16, 1972), pp.
of conditionalprobability rank as follows' (1) Choice of source 11-16, U.S. Department of Housing and Urban Development,
zone, varies T• by 100% and varies Pc by ,• 50%, (2)choice of Washington,D.C., 1972.
line-fitting technique, varies T• by 50-70% and varies Pc by Algermissen,S. T., D. M. Perkins,P. C. Thenhaus,S. L. Hanson, and
30-40%, (3) choice of standard deviation, varies Pc by 10- B. L. Bender, Probabilistic estimatesof maximum ground acceler-
ation and velocity in rock in the contiguousUnited States, U.S.
25%, and (4) choiceof distribution, varies Pc by ~ 10%. Items Geol.Surv.OpenFile Rep.,82-1033,99 pp., 1982.
1 and 2 above required a choice be made in order to derive Applied Technology Council, Tentative Provisionsfor the Devel-
final probability values. Arguments were presentedfor select- opmentof SeismicRequlationsfor Buildinqs,Publ. 78-8, 505 pp.,
ing 1, the large sourcezone and 2, least squaresregression,for Natl. Sci. Found., Washington,D.C., 1978.
the final estimates.The range of the final estimatesare due to Bender,B., Maximum likelihood estimationof b valuesfor magnitude
groupeddata, Bull. Seismol.Soc.Am., 73, 831-851, 1983.
items 3 and 4.
Braile, L. W., G. R. Keller, J. W. Hinze, and E.G. Lidiak, An ancient
We have been careful to make clear the number of assump- rift complexand its relation to contemporaryseismicityin the New
tions necessaryto derive the above probability values. The Madrid seismiczone, Tectonics,1, 225-237, 1982.
critical data required to reduce or constrain these estimates Brillinger,D. R., Seismicrisk assessment: Somestatisticalaspects,
would be event chronologiesbased on palcoseismicevidence. Earthq•uake Predict.Res.,1, 183-195,1982.
Bufe, C. G., P. W. Harsh, and R. O. Burford, Steady-stateseismic
Broad zonesof the embaymentsedimentsoverlyingthe New slip•A:"preciserecurrence
model,Geophys.
Res.Lett.,4, 91-94,
Madridseismic
zonehavehighliquifaction
potential.
Datable 1977.

sandblows,slumps,or other liquifactionfeaturesproducedby Cluff, L. S., A. S. Patwardhan,and K. J. Coppersmith,Estimatingthe

pre-1811 seismicepisodesand preserved•in thesesediments probability of occurrencesof surfacefaulting earthquakeson the
Wasatch fault zone, Utah, Bull. Seismol.Soc. Am., 70, 1463-1478,
offer the best hope of obtaining more tightly constrainedfore- 1980. •
casts of the future seismic behavior of the New Madrid zone. Ervin, C. P., and L. D. McGinnis,Reetfootrift: Reactivatedprecursor
to the Mississippiembayment,Geol.Soc.Am. Bull.,86, 1287,1295,
1975.

TABLE 7. Final Probability Estimates Hagiwara, Y., Probability of earthquakeoccurrenceas obtainedfrom

a Weibull distributionanalysisof crustalstrain, Tectonophysics,
23,
Average Probability Probability 313-318, 1974.
Repeat by Year by Year Harada, T., Quiet and violen• in horizontal movementof the crust,
Time, 2000 A.D., 2035 A.D., Earthquake Precursors,edited by C. Kisslingerand Z. Suzuki, •.
years % % Phys.Earth, suppl.issue,79-83, 1978.
Howell, B. F., A comparisonof estimatesof seismicrisk in the central
r% >_6.0(M• >_6.3) 70 (4-15) 40-63 86-97 United States,EarthquakeNotes,51, 13-19, 1980.
r% >_6.6(M• _>7.6) 254 ( + 60) 5.4-8.7 19-29 Jacob, K. H., Estimates of long-term probabilities for future great
r% >_7.0(M, >_8.3) 550 (4-125) 0.3-1.0 2.7-4.0 earthquakes in the Aleutians, Geophys.Res. Lett., 11, 295-298,
1984.
 JOHNSTON AND NAVA.' NEW MADRID
Johnston,A. C., On the use of the frequency-magnituderelation in
earthquake risk assessment, in Earthquakes and Earthquake
Engineering-EasternUnited States, vol. 1, edited by J. E. Beavers,
pp. 161-181, Ann Arbor Science,Ann Arbor, Mich., 1981.
Kiremidjian, A. S., and T. Anagnos,Stochasticslip-predictablemodel
for earthquake occurrences,Bull. Seismol.Soc. Am., 74, 739-755,
1984.
Lomnitz, C., Global Tectonicsand Earthquake Risk, 320 pp., Elsevier
Scientific, Amsterdam, Netherlands, 1974.
Mann, O. C., W. Howe, and F. H. Kellogg, Regional earthquake risk
study, technicalreport, 398 pp., Dep. of Housing and Urban Dev.,
Washington, D.C., 1974.
McClain, W. C., and O. H. Myers, Seismichistory and seismicityof
the southeastern region of the United States, Rep. ORNL-4582,
Oak Ridge Natl. Lab., Oak Ridge, Tenn., 1970.
McNally, K. C., and J. B. Minister, Nonuniform seismicslip along the
Middle American Trench, J. Geophys.Res.,86, 4949-4959, 1981.
Miller, A. R., FORTRAN Programsfor Scientists and Engineers, 280
pp., SYBEX, Berkeley,Calif., 1982.
National Bureau of Standards, Tables of Normal Probability Func-
tions, NBS Appl. Math. Set., vol. 23, 344 pp., U.S. Government
Printing Office, Washington,D.C., 1953.
Nishenko, S. P., Seismicpotential for large and great interplate earth-
quakesalong the Chilean and southernPeruvian margins of South
America: A quantitative reappraisal, J. Geophys. Res, 90, 3589-
3615, 1985.
Nuttli, O. W., Magnitude-recurrence relation for central Mississippi
Valley earthquakes,Bull. Seismol.Soc.Am.,64, 1189-1207,1974.
Nuttli, O. W., Seismicityof the centralUnited States,Geology in the
Sitingof Nuclear Power Plants,Rev.Eng. Geol.,4, 67-93, 1979.
Nuttli, O. W., Seismichazard associatedwith the New Madrid fault
zone, Evaluation of Past Studiesand Identification of Needed Stud-
ies of the Effects of Major Earthquakes Occurring in the New
Madrid Seismic Zone, Appendix A, 15 pp., Fed. Emergency
Manage. Agency,KansasCity, MO., 1981.
Nuttli, O. W., Average seismicsource-parameterrelations for mid-
plateearthquakes,Bull. Seismol.Soc.Am., 73, 519-535, 1983.
Nuttli, O. W., and K. G. Brill, Earthquakesourcezonesin the central
United Statesdeterminedfrom historical seismicity,An Approach
to SeismicZonation for Siting Nuclear Electric Power Generating
Facilities in the Eastern United States,Rep. NUREG/CR-1577, pp.
98-142, Nucl. Regul.Comm., Washington,D.C., 1981.
Nuttli, O. W., and R. B. Herrmann, Credible earthquakes for the
central United States, State-of-the-art for AssessingEarthquake
Hazards in the United States,Misc. Pap. 5-73-1, Rep. 12, 99 pp.,
U.S. Army Corps of Eng., Waterways Exp. Stn., Vicksburg,Miss.,
1978.
Obermeier, S. F., G. S. Gohn, R. E. Weems, R. L. Gelinas, and M.
Rubin, Geologic evidencefor moderate to large earthquakesnear
Charleston, South Carolina, Science,227, 408-411, 1985.
Perry, R. G., Seismichazard analysisfor the central United States,
M.S. thesis,175 pp., Saint Louis Univ., Saint Louis, MO., 1981.
Raleigh,C. B., K. Sieh,L. R. Sykes,and D. L. Anderson,Forecasting
southernCalifornia earthquakes,Science,217, 1097-1104, 1982.
Rikitake, T., Statisticsof ultimate strain of the earth's crust and prob-
ability of earthquakeoccurrence,Tectonophysics, 23, 1-21, 1975.
Rikitake, T., Possibleprocedure of earthquake prediction and some
problemsof earthquakewarning(in Japanese),in A Symposium on
EarthquakePredictionResearch,edited by Z. Suzuki and S. Omote,
pp. 215-224, SeismologicalSocietyof Japan,Tokyo, 1977.
PROBABILITIES 6753
Rikitake, T., Practical approachto earthquakepredictionand warn-
ing, in CurrentResearchin EarthquakePrediction1, vol. 2, edited
by T. Rikitake,pp. 1-52, D. Reidel,Hingham,Mass.,1981.
Russ,D. P., Late Holocenefaulting and earthquakerecurrencein the
Reelfoot Lake area, northwestern Tennessee,Geol. Soc. Am. Bull.
90(Part 1), 1013-1018, 1979.
Shi, Y., and B. A. Bolt, The standard error of the magnitude fre-
quencyb value,Bull.Seismol.
Soc.Am.,72, 1677-1687,1982.
Shimazaki, K., and T. Nakata, Time-predictablerecurrencemodel for
largeearthquakes,Geophys. Res.Lett., 7, 279-282, 1980.
Sieh,K. E., Prehistoriclargeearthquakes
producedby slipon the San
Andreas fault at Pallett Creek, California, J. Geophys.Res., 83,
3907-3939, 1978.
Stauder,W., Present-dayseismicityand identificationof activefaults
in the New Madrid seismiczone, Investigationof the New Madrid,
Missouri,EarthquakeRegion, U.S. Geol.Surv.Prof. Pap., 1236C,
21-30, 1982.
Stauder, W., et al., Central MississippiValley Earthquake Bulletin,
1-36, Saint Louis Univ., Saint Louis, MO., 1974-1983.
Stauder, W., G. Fischer, S. Schaeffer,and S. T. Morrisey, Seismic
characteristicsof southeast Missouri as indicated by a regional
telemeteredmicroearthquakearray, Bull. Seismol.Soc. Am., 66,
1953-1964, 1976.
Swan, F. H., III, D. P. Schwartz, and L. S. Cluff, Recurrenceof
moderate to large magnitude earthquakesproduced by surface
faulting on the Wasatchfault zone, Utah, Bull. Seismol.Soc.Am.,
70, 1431-1462, 1980.
Sykes,L. R., and S. P. Nishenko,Probabilitiesof occurrenceof large
plate rupturingearthquakesfor the San Andreas,San Jacinto,and
Imperial faults,California, 1983-2003,J. Geophys.Res.,89, 5905-
5927, 1984.
Till, R., Statistical Methods for the Earth Scientist, 154 pp., John
Wiley, New York, 1974.
Tsuboi,C., Earthquakeenergy,earthquakevolume,aftershockarea,
and strengthof the earth's crust, J. Phys. Earthquake,4, 63-66,
1956.
Weibull, W., A statisticaldistribution function of wide application, J.
Appl. Mech., 18, 293-297, 1951.
Weichert, D. H., Estimation of the earthquake recurrenceparameters
for unequalobservationperiodsfor differentmagnitudes,
Bull.Seis-
tool. Soc. Am., 70, 1337-1346, 1980.
Wesnousky,S. G., C. H. Scholz,and K. Shimazaki,Deformationof
an island arc: Rates of moment releaseand crustal shortening in
intraplateJapandeterminedfrom seismicityand Quaternaryfault
data, J. Geophys.Res.,87, 6829-6852, 1982.
Wesnousky,S. G., C. H. Scholz,K. Shimazaki,and T. Matsuda,
Earthquake frequencydistributionand mechanicsof faulting, J.
Geophys.Res.,88, 9331-9340, 1983.
Wesnousky,S. G., C. H. Scholz, K. Shimazaki, and T. Matsuda,
Integrationof geologicaland seismological
data for the analysisof
seismichazard: A casestudy of Japan, Bull. Seismol.Soc. Am., 74,
687-708, 1984.
A. C. Johnston and S. J. Nava, TennesseeEarthquake Information
Center,MemphisStateUniversity,Memphis,TN 38152.
(ReceivedOctober 29, 1984;
revised March 12, 1985;
acceptedMarch 12, 1985.)